The electrocardiogram F-wave arising from atrial electrical activity is an important global measure for assessment of atrial fibrillation (AF). However, successful F-wave extraction from the ventricular waveform can be problematic. Herein, a new F-wave isolation technique is introduced. For analysis, electrocardiogram lead I (termed unfiltered or UNF-signals) was retrospectively analyzed (39 AF patients, 8.4-s recordings, 8192 sample points, 96 recordings in total). To measure the efficacy of isolation techniques, a synthetic F-wave (7.29 Hz) and an interference were added to each electrocardiogram signal. In the resulting composite signals, the average electrocardiogram QRST complex template was subtracted from each actual QRST (AVG-isolation). The QRST template was also adjusted using a new adaptive least mean-squares (LMS) algorithm, and subtracted from each actual QRST (termed LMS-isolation). Four spectral parameters were measured to assess isolated F-wave quality: the dominant amplitude (DA), dominant frequency (DF) and mean/standard deviation in spectral profile (MP/SP). Significant parameter differences between UNF/LMS and between AVG/LMS were determined. The electrocardiogram F-wave spectral parameters were significantly improved by incorporating LMS-isolation as compared to no isolation (p < 0.001). The F-wave spectral parameters were also significantly improved using LMS-isolation as compared with AVG-isolation (DA/MP/SP: p < 0.001; DF: p < 0.05). The DF was correctly identified as 7.29 ± 0.10 Hz using ensemble spectral analysis with the following percentages (UNF: 24.0%, AVG: 69.8%, LMS: 80.2%), and Fourier spectral analysis with the following percentages (UNF: 15.6%, AVG: 60.4%, LMS: 75.0%). The LMS algorithm is helpful to isolate the electrocardiogram F-wave from the ventricular component as measured by spectral analysis, when compared to the use of an average QRST subtraction template.
High-Selectivity Filter Banks for Spectral Analysis of Music Signals
